AI Infrastructure Agent
Type: full-code · Vendor: VersusControl · Language: Go · License: Apache-2.0 · Status: active · Status in practice: experimental · First released: 2025-09-01
AI Infrastructure Agent manages AWS infrastructure from natural-language requests by planning the changes, presenting a dry-run plan for human approval, and only then applying them.
Description. AI Infrastructure Agent is a Go tool that lets a user manage AWS infrastructure through natural-language commands. It is powered by a configurable model provider (OpenAI, Google Gemini, or Anthropic Claude) and exposes capabilities over the Model Context Protocol. The agent analyzes a request, reads the current infrastructure state, and creates a detailed execution plan; every operation can first run in dry-run mode that shows exactly what would be created, modified, or deleted. The agent presents that plan for review and waits for explicit approval before any real AWS changes are applied.
Agent loop shape. The agent receives a natural-language request, queries available tools and the current infrastructure state, and uses a configured model to create a detailed execution plan. Operations can run in dry-run mode that shows exactly what would be created, modified, or deleted. The plan is presented for the user's review and the agent waits for approval before executing any real change, handling resource dependencies automatically during apply.
Primary use cases
- natural-language AWS infrastructure management
- dry-run preview of infrastructure changes before apply
- human-approved provisioning of cloud resources
- state-aware planning over existing AWS resources
Key concepts
- Dry-run mode → simulate-before-actuate (docs) — A preview mode in which every operation is simulated and the agent reports exactly what would be created, modified, or deleted, and estimates cost, without touching any AWS resource.
- Execution plan → plan-and-execute (docs) — The model-generated, ordered set of infrastructure operations the agent proposes for a request; it is shown for review and applied only after approval, with resources created in dependency order.
- Terraform-like state (docs) — An internally maintained record of the current infrastructure state that the agent reads before planning and uses to detect drift from the expected configuration and resolve conflicts.
- MCP Server (Agent Core) → mcp (docs) — The Go core that implements the Model Context Protocol, exposes the infrastructure tools, and runs the decision-making and planning loop behind the React web dashboard.
Patterns this full-code implements —
- ★★Plan-and-Execute
The agent first generates a complete execution plan with the configured model from the request and current state, and only once approved executes the planned operations in order, separating a plannin…
- ★★Model Context Protocol
The core agent is implemented as an MCP server, so its infrastructure capabilities are exposed and invoked through the Model Context Protocol.
- ★★Tool Use
Before planning, the agent queries the MCP server for the available tools and their capabilities and selects from them to assemble the execution plan.
- ★Simulate Before Actuate
The agent runs every infrastructure operation in a dry-run mode that shows exactly what would be created/modified/deleted and presents the execution plan for human approval before any real AWS change…
- ★★Human-in-the-Loop
The agent analyzes the request, creates a detailed execution plan, presents it for the user's review showing exactly what will be created, and waits for explicit approval before applying any change.
- ★Synchronous Execution-Plan Confirmation
The agent emits its full execution plan and blocks on the user's confirmation before any side-effecting step, so apply only proceeds after the plan is reviewed and approved.
Neighbourhood
Click any neighbour to follow the lineage. Scroll to zoom, drag to pan.
Alternatives & relatives
- full-code · framework
Amazon Bedrock Agents
Amazon Bedrock is one of the configurable model providers for AI Infrastructure Agent (AWS Bedrock Nova), and Bedrock Agents is AWS's own managed agent runtime over the same cloud.
- app · closed product
Kiro
Both keep a human approval gate over a model-proposed plan, but Kiro is AWS's spec-driven coding agent for application code while this agent plans and applies AWS infrastructure changes.
References
Provenance
- Last analyzed:
- Last updated:
- Verification status: partial